When Robots Learn to Belong: How ‘Embodied’ AI Is Shaping Society’s Next Chapter
Imagine a robot not simply following programmed instructions, but genuinely responding to your cues—walking alongside you, picking up objects, knowing when your smile means “thanks” or when your frown means “that wasn’t right.” That scenario, once the stuff of science-fiction, is fast becoming part of today’s dialogue. In a recent piece for the World Economic Forum, researcher Vanessa Evers explores how artificial intelligence (AI) in physical form—so-called “embodied AI”—is poised to join our homes, workplaces and streets, and what it will mean for society at large. ([World Economic Forum][1])
Why this matters
Current AI systems—like chatbots and large-language models (LLMs)—excel at processing and generating text. But giving AI a body changes everything. As Evers explains: “What happens when AI gains a body?” ([World Economic Forum][1]) Robots working in the physical world face vastly more complexity—doors, windows, walls, crowds, unexpected events—not just words. ([World Economic Forum][1])
For example: A robot standing in front of a front door must understand that the door swings on hinges, the bell is fixed, walls don’t vanish. It needs a grounded “model of the world”—physics, objects, social norms. ([World Economic Forum][1]) In contrast, LLMs are trained on trillions of words from the internet. ([World Economic Forum][1])
What humans teach us about robots
One of the article’s key insights: humans don’t just learn from data—they learn from bodies, from culture, from social interaction. Our children absorb huge amounts of sensory input (vision, audio, touch) in the first years of life. ([World Economic Forum][1]) Humans learn what philosopher Daniel Dennett called enculturation—we internalize norms, behaviours, values by observing, being taught, interacting. ([World Economic Forum][1])
By contrast, robots today rarely learn in this richly embodied, socially contextualised way. The article argues we need to combine:
- Social Reinforcement Learning (SRL): robots detecting social signals (smile, frown, posture) and adjusting behaviour accordingly. ([World Economic Forum][1])
- Active Teaching: humans guiding robots by demonstration or by choosing the right action among options, in real or simulated environments. ([World Economic Forum][1])
The challenge: Hierarchical planning and unpredictability
Even if a robot can interpret social signals and be taught, it still has to navigate complexity. For instance: take someone to the train station. That sounds simple—but it involves sub-goals (leaving the apartment, locking the door, taking a bus, platforms, changing trains) each with unpredictability. ([World Economic Forum][1])
Robots can’t script every minor action. They need to combine pre-training with adaptability: planning, sub-goal switching, sensing humans and the environment in real time. Active human teaching may be crucial here. ([World Economic Forum][1])
Why “now” is the moment
We’re already seeing robots beyond assembly lines: delivering food in cities, supporting logistics, providing companionship in care homes. ([World Economic Forum][1]) With AI entering physical machines, the stakes multiply:
- Opportunities: Robots can undertake dangerous tasks (construction, mining), assist elderly care, respond to disasters. ([World Economic Forum][1])
- Risks: They could displace jobs, exacerbate inequality, raise privacy/safety/trust questions, or even be weaponised. ([World Economic Forum][1])
- Choices: Policymakers, designers, technologists must decide: where do robots belong? Who benefits? How do we build trust and inclusive systems? ([World Economic Forum][1])
A Proposal: “Cultural Robotics”
The article introduces the concept of cultural robotics—training robots not just on data, but through embodied interaction, social cues and active human teaching. The suggestion: robots must be enculturated if they’re to share our lives—not just perform tasks. ([World Economic Forum][1])
In other words: The question isn’t if AI will get a body, but how we will teach it to belong in our world. ([World Economic Forum][1])
Why it’s worth paying attention
This discussion is foundational for how society might integrate robots widely. It reframes robots not just as machines doing tasks, but as social participants. The shift demands rethinking how we design, teach, regulate and deploy embodied AI. For companies, governments, citizens alike, the transition has concrete implications: workforce change, regulation, ethics, everyday human-robot interaction.
Glossary
- Embodied AI: Artificial intelligence integrated into a physical body (robot) that acts in the physical world, not just processing text or data.
- Large Language Model (LLM): An AI model trained on vast quantities of text to predict the next word or sequence, underpinning tools like chatbots.
- Enculturation: The process by which an individual learns and adopts the behaviours, values, norms of their culture—used here as a metaphor for how robots might learn human-social norms.
- Social Reinforcement Learning (SRL): A learning method where an agent (robot) uses social feedback signals (smiles, frowns, gestures) to guide behaviour.
- Active Teaching: A learning paradigm where a human actively demonstrates, selects or guides actions of an AI/robot to shape its behaviour.
- Hierarchical Planning: A planning strategy that breaks a complex goal into multiple sub-goals, each requiring decision and adaptation, especially relevant for robots navigating real environments.
Conclusion
As AI shifts from keyboards and screens into our physical space, the process of making robots fit in becomes less about raw data and more about culture, interaction and trust. The article from the World Economic Forum highlights that to truly belong in human society, robots need to learn from humans—not just about humans. The journey ahead isn’t purely technological, it’s deeply human.
Source: https://www.weforum.org/stories/2025/10/ai-robots-learn-to-belong-human-society/
| [1]: https://www.weforum.org/stories/2025/10/ai-robots-learn-to-belong-human-society/ “How AI-driven robots can learn to belong in human society | World Economic Forum” |
